As AI workloads keep moving closer to the edge, storage is quietly becoming one of the real bottlenecks. GPUs get all the attention, but anyone who has actually tried to run large local models, fine-tune data, or build AI tools on a desktop knows that slow storage can bring everything to a crawl. Samsung’s PM9E1 SSD is built for exactly that reality, and it is one of the more interesting pieces of hardware to land alongside NVIDIA’s new DGX Spark desktop AI system.
At a glance, the PM9E1 already stands out because of its size. This is a PCIe Gen5 SSD that fits into an M.2 22×42 form factor, which is much shorter than the more common 2280 drives. That matters in compact AI systems where every millimeter counts and where storage has to sit close to compute to keep latency down. Despite the smaller footprint, Samsung managed to pack up to 4TB into this tiny module, which is not trivial when you factor in thermals, power, and sustained performance.
Performance is where the PM9E1 really earns its keep. Sequential read speeds reach up to 14,500MB per second, while writes hit up to 12,600MB per second. Random performance is just as important for AI workloads, and here the drive pushes up to 2,000K IOPS for reads and 2,640K IOPS for writes. Those numbers are not just for spec sheets. They translate into faster dataset loading, quicker model checkpoints, and less waiting when you are moving between experiments.
SEE ALSO: MSI EdgeXpert launches as another NVIDIA DGX Spark clone for local AI
Samsung is also making a big deal about efficiency, and for once it is not just marketing fluff. The PM9E1 is designed to deliver roughly double the performance of the previous generation while improving power efficiency by up to 45 percent. In a desktop AI system that is already packed with heat-generating components, that efficiency directly affects stability, noise, and long-term reliability. Less wasted power also means more flexibility in system design, which is part of how something like DGX Spark can exist as a desktop machine instead of a rack-mounted monster.
What makes the PM9E1 genuinely AI-focused is the controller. Samsung uses its in-house Presto controller, built on a 5nm process, with firmware tuned specifically for AI software stacks. It is optimized for DGX Spark OS, NVIDIA CUDA, and AI-heavy workflows that mix large sequential transfers with bursts of random access. That combination is important because AI storage is not just about moving huge files. It is about doing that while also responding instantly to small, frequent reads and writes. The PM9E1 handles both without falling apart under load.
Unlike some consumer drives that cut corners, the PM9E1 includes dedicated DRAM, even in this compact form factor. That helps keep performance consistent instead of collapsing once caches fill up. For developers, researchers, and creators running long jobs, consistency matters more than peak numbers. Nobody wants a drive that looks fast for the first minute and then slows to a crawl halfway through a training run.
SEE ALSO: PNY unveils slim NVIDIA RTX 50 GPUs built for tight PC cases
Security is another area where Samsung is clearly thinking about AI systems as more than just fast PCs. The PM9E1 supports SPDM version 1.2, which enables device authentication, firmware attestation, and secure communication channels. This brings certificate-based security down to the component level, which is especially important when AI systems are handling proprietary data, models, or sensitive workloads. It also avoids the irony of burning AI compute just to verify storage integrity.
The PM9E1 is already qualified and in mass production for use in NVIDIA DGX Spark, which says a lot about its maturity. NVIDIA does not usually ship experimental storage in a flagship platform, especially one meant to sit on desks rather than in data centers. The combination of performance, capacity, efficiency, and security is what makes this drive a practical choice rather than a flashy prototype.
In the bigger picture, the PM9E1 shows where local AI computing is heading. As models shrink, improve, and move onto desktops, laptops, and small workstations, storage has to keep up. Fast GPUs without equally fast storage just leave performance on the table. Samsung’s approach here feels very deliberate, focusing on balance instead of chasing a single headline metric.
As an Amazon Associate I earn from qualifying purchases. Some links in this article may be affiliate links. If you click one and make a purchase, I may receive a small commission at no extra cost to you. This helps support NERDS.xyz.
